meta: commit new chunks in write order

[davies]

close #5038

[Copilot]

Pull request overview

This PR aims to prevent observed “zero-filled regions” during concurrent reads of a file that is actively being appended to, by enforcing a commit ordering constraint across newly growing chunks in pkg/vfs (close #5038).

Changes:

• Track whether a slice contributes to file growth (growing) and whether it has been committed (committed).
• Introduce cross-chunk commit dependencies so a new chunk’s first slice waits for the prior growing slice to commit before writing metadata.
• Add a per-file condition variable (commitcond) to coordinate commit ordering across chunk commit goroutines.

Comments suppressed due to low confidence (1)

pkg/vfs/writer.go:307

• Finding the dependency for a new chunk scans the entire f.chunks map (and then scans lastChunk.slices) while holding the file lock. With many buffered chunks/slices this can become O(n²) during sequential appends. Consider tracking the last in-progress/growing slice (or last uncommitted chunk index) on fileWriter and updating it when s.growing is set/committed, so setting s.dep is O(1). Also, the inner loop variable c shadows the outer c (chunkWriter), which makes this block harder to follow; renaming the inner variable would improve readability.

			if uint64(indx)*meta.ChunkSize >= f.length {
				// first slice of a new chunk, try to find the last slice of the last chunk as dependency
				var lastChunk *chunkWriter
				for i, c := range f.chunks {
					if i < indx && (lastChunk == nil || i > lastChunk.indx) {
						lastChunk = c
					}
				}
				if lastChunk != nil {
					var lastSlice *sliceWriter
					for _, s := range lastChunk.slices {
						if s.growing {
							lastSlice = s
						}
					}
					s.dep = lastSlice
				}
			}

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.